Postgres Node Returns Empty Result, but Query Works in pgAdmin

[avsion]

Bug Description

"Hello, I have an n8n workflow with a Postgres node that always returns an empty result [ ].

I have confirmed the exact same SQL query works perfectly when run directly in pgAdmin with the same user, and it finds the data.
I have confirmed my n8n credential points to the exact same host, database, and user as my working pgAdmin connection.
I have already restarted the n8n and Postgres containers multiple times.
I have tried different query formats (ILIKE and lower()).
This seems to be an environmental issue with how the n8n node is connecting. Can anyone help me troubleshoot what might be happening at the connection/environment level?"

To Reproduce

I have confirmed the exact same SQL query works perfectly when run directly in pgAdmin with the same user, and it finds the data.
I have confirmed my n8n credential points to the exact same host, database, and user as my working pgAdmin connection.
I have already restarted the n8n and Postgres containers multiple times.
I have tried different query formats (ILIKE and lower()).
This seems to be an environmental issue with how the n8n node is connecting. Can anyone help me troubleshoot what might be happening at the connection/environment level?"

Expected behavior

output contact name from the database table or even in manual test

Operating System

Ubuntu 204

n8n Version

1.98.2

Node.js Version

v22.16.0

Database

PostgreSQL

Execution mode

main (default)

[Joffcom]

Hey @avsion,

We have created an internal ticket to look into this which we will be tracking as "GHC-2662"

[Joffcom]

Hey @avsion,

Do you have any screenshots of the input to the node and the output? Normally we see this when expressions are being used and the value is empty.

[avsion]

Here you go

[Joffcom]

It might be just me but that query looks different, What happens if you also add in the lowers and use ILIKE, I wouldn't expect that to change anything but lets make sure we are comparing the same thing.

We also expect a comma seperated list for the parameters not an array, What happens if you just use Rachel

[avsion]

Hello,

Both test below output came empty

Test 1: Combined Query

Keep the Query Parameters field set to [ "Rachel" ].
Update the Query field to this version that uses both lower() and ILIKE:

SQL

SELECT *
FROM public.contacts
WHERE lower(first_name) ILIKE lower('%' || $1 || '%') OR lower(last_name) ILIKE lower('%' || $1 || '%');

Execute the node and let us know if the output is still empty.

Test 2: Plain Text Parameter

If Test 1 still results in an empty output:

Keep the same query from Test 1.
Change the Query Parameters field from [ "Rachel" ] to be just the plain text, without any brackets or quotes: Rachel
Execute the node and let us know what the result is (even if it's a new error).

[Joffcom]

I have tried 2 options now...

1. The Postgres Node as a standalone app node

2. The Postgres Node as an AI tool using OpenAI as the model

In both cases this has worked, One thing I have noticed is in your screenshots the node has not actually executed at all which is why the output page is empty and doesn't show any kind of response. Even with data that doesn't exist it should still return something like this.

I think the issue here might be that you are either using an AI model that doesn't work with tools very well (or Gemini), This can be tested by trying another model or your prompt is not set to call the tool so it isn't bothering.

As one last test can you try using the Postgres App node instead of the Agent tool and put in the same query and let me know if that works, As both nodes do the same thing internally it will also help confirm if this is an issue with the model or not.

[avsion]

Hello Joffcom,

Thank you for the suggestions. I have performed the tests you recommended and have a much more detailed analysis of the issue now.

When executing the Find_Contact node manually in the editor, both of your suggested tests (the combined lower()/ILIKE query and the plain text parameter) still resulted in a failed state or an empty output.

However, the key findings come from workflow tests triggered from Telegram. Here is the exact sequence of events I've discovered:

The AI Agent receives the user's prompt correctly (e.g., "Find Rachel's contact").
The AI model and prompt are working perfectly, because the agent correctly decides to use the Find_Contact tool to fulfill the request.
However, the Find_Contact tool node itself never receives any input data. The execution log shows "No items were sent on this branch" for its input.
The output of the AI Agent node itself is not a tool call, but a final, generated text response. The AI composes an apology message, such as: "I apologize, but I was unable to find Rachel in your contacts. There was an error when trying to access the contact database..."
Conclusion:

This proves that the AI Agent is internally trying to execute the Find_Contact tool, but the tool is failing during that internal process. The AI Agent is robust enough to catch this failure, abort the tool call, and generate a user-friendly apology instead of crashing the entire workflow.

This isolates the problem definitively. It is not an issue with a specific AI model or the prompt—I have confirmed this exact failure sequence occurs when using both Google's Gemini and OpenAI models as the agent's brain.

The issue is a fundamental, environmental problem where the Postgres node fails to execute correctly only when called by the AI Agent, even though the connection credentials are correct and the same query works in pgAdmin.

[Joffcom]

Hey @avsion,

No input data yet does not mean the node executed, If you open a node without running it you will see the same thing. It looks like the node probably failed to even start because of your expression, The Postgres node doesn't have a parameter called name. The $parameter option is for using parameters set in the node not for parsing incoming options.

Try using {{ $fromAI('name', The name from the question, 'string') }} instead and don't add the [ ] this should then work if you were to ask something like... "What are the details for Rachel".